Best Starting Point: Coral USB for Detection, Intel iGPU for Video, No Discrete GPU Yet
For a six-camera Frigate build on ZimaBoard 2 1664, start simple. Use the Coral USB for object detection and the Intel N150 iGPU for video decode/encode through VAAPI or Quick Sync. Add a discrete GPU only if measured workload proves the iGPU is the bottleneck.
The ZimaBoard 2 product page confirms the Intel N150 platform, 16 GB option, dual SATA and PCIe expansion. That gives you room to add hardware later without paying the power and PCIe cost up front.
Why This Split Makes Sense in Frigate
Frigate treats object detection and video decoding as separate acceleration jobs. Its configuration guide includes a working pattern using a USB Coral for detection and VAAPI for video. Its video acceleration documentation covers Intel VAAPI/QSV presets.
A Single USB Coral Is Often Enough for Six Cameras
Frigate’s current hardware guidance says one Coral can handle many cameras with the default model, and recommends looking at actual inference speed before adding a second detector. Camera count alone does not justify a second accelerator.
For Storage, Separate Recording I/O From the System When You Can
Continuous CCTV writes are a different workload from ZimaOS app data. A practical layout is system/app data on SSD or eMMC and recordings on a dedicated HDD. RAID 1 is useful if you value drive-failure continuity, but it halves usable capacity and it is not a backup.
The current ZimaOS backup guide makes that distinction explicit. For broader app sizing, see the ZimaOS App Store hardware guide.
When a Discrete GPU Becomes Worth It
Add one only after you can name the bottleneck: decode load, encode/restream load, enrichments or another GPU workload sharing the box. Buying a GPU “just in case” is the least efficient part of this build.
